On Mar 5, 9:49 am, m...@reiss.demon.co.uk wrote:   
   > > you wrote in page: the task of each output note was to predict the   
   > > activity of its corresponding input node   
   >   
   > > what exactly does this mean, given multiple input nodes?   
   >   
   > I'm not sure exactly what you're confused about - so I shall spell   
   > out the whole thing as carefully as I can.   
   >   
   > The input and output layers of nodes are two dimensional arrays   
   > of the same size. So there is a simple topological mapping from an   
   > input node to its "corresponding" output node. For example, take the   
   > node in the bottom left of the input array - the "corresponding"   
   > output node would be in the bottom left of the output array.   
   >   
   > Each output node is given the task of predicting the activity of its   
   > "corresponding" input node. That is to say it should be active when   
   > the corresponding input node is active. But it is important to note   
   > that there is no direct connection between the corresponding input   
   > and output nodes, so the output has to infer the activity of its   
   > corresponding input node from the signals it gets from the other   
   > nodes it is connected to.   
   >   
   > I guess some confusion may have been caused by the phrase   
   > "corresponding input node" which may accidentally imply that it feeds   
   > an   
   > input to the output node (which it doesn't). Maybe I should use some   
   > other expression like "topologically mapped node in the input layer".   
   >   
   > Does that answer your question?   
   >   
   > M.   
   >   
      
   so what you mean is 2nd layer is made of nodes that should predict   
   value of non-connected 1st layer nodes in moment T given signals from   
   connected 1st layer nodes in moment(s) T-1 (-2, -3, ...)?   
      
   I can imagine this being useful for noise reduction and otherwise   
   automatically patching images (or generally, interpolating partial   
   data), but beyond that - in context of cortex immitation - what else   
   it is supposed to do?   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)